ivakegg commented on pull request #1656:
URL: https://github.com/apache/accumulo/pull/1656#issuecomment-795906094


   > @friedlou What does this change accomplish? There is no description of the 
change that helps us understand what it is trying to accomplish. From what I 
can tell, the most obvious consequence is fewer log messages: one for all 
failures, rather than individual messages for each failure in the loop that is 
skipped later, perhaps? I'm not sure that's much of an improvement, especially 
since, as @milleruntime says, it only affects the legacy bulk import. Am I 
missing something? Is there another benefit to this change?
   
   The main reason for this is that we have timing issues associated with bulk 
loading where we need to ensure that the master waits long enough for the 
tserver to finish its retries. We have multiple situations that occur when 
tservers fail to assign things in a timely fashion.  The most nasty of those is 
where the garbage collector has removed the file because all of the other 
tablets have successfully assigned and major compacted away.  Meanwhile a 
tablet doing retries finally assigns the file and finds it gone.  When we have 
retries if becomes difficult to bound the time so we were attempting to set the 
retries to 0 only to find out that it still retried once.


----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
[email protected]


Reply via email to